In premixed turbulent combustion models, two mechanisms have been used to explain the increase in the flame speed due to the turbulence. The newer explanation considers the full range of turbulence scales which wrinkle the flame front so as to increase the flame front area and, thus, the flame propagation speed. The fractal combustion model is an example of this concept. The older mechanism assumes that turbulence enables the penetration of unburned mixtures across the flame front via entrainment into the burned mixture zone. The entrainment combustion or eddy burning model is an example of this mechanism. The results of experimental studies of combustion regimes and the flame structures in SI engines has confirmed that most combustion takes place at the wrinkled flame front with additional combustion taking place in the form of flame fingers or peninsulas. As the ratio of the turbulence intensity to the laminar flame speed increases, the importance of the flame peninsulas should become increasingly important. While it has been shown that fractal geometry can be used to account for flame wrinkling, it may be difficult to extend this concept to account for the additional surface area resulting from the flame peninsulas. However, the flame front convolution that results in flame peninsulas can be envisioned as entrainment combustion. In the present research an effort was made to combine the fractal combustion model and the entrainment combustion model to generate the “peninsula-fractal” combustion model, so as to improve burn rate predictions of SI engine codes.Two engines, a 2-valve and a 4-valve, were used to validate the new combustion model. Comparisons of the original engine code (with the entrainment combustion model modified to account for the effects of flame strain), the new code (with the peninsula-fractal combustion model) and engine test data show that the burn rate predictions for the fully developed phase of combustion (10-90% mass fraction burned or MFB) with the new model are significantly improved (the original code adequately predicted 0-2% MFB and 0-10% MFB). The new model calculations of 10-90% MFB durations are within 2 CA degrees of the test data for both engines and over a range of speeds, loads, air/fuel ratios, and EGR rates. This is considered good for engine codes. In contrast, the old model results are off by 4 to 6 CA degrees compared with test data.